BindingSite-AugmentedDTA: enabling a next-generation pipeline for interpretable prediction models in drug repurposing

Area of Science:

• Computational Chemistry
• Bioinformatics
• Drug Discovery

Background:

• Drug-target interaction (DTI) prediction is crucial for drug discovery.
• Existing methods often lack generalizability and interpretability.
• Accurate prediction of drug-target affinity (DTA) is essential for efficient drug development.

Purpose of the Study:

• To propose a deep learning (DL) framework, BindingSite-AugmentedDTA, to improve DTA predictions.
• To enhance the generalizability and interpretability of DTA prediction models.
• To reduce the search space for protein binding sites for more efficient and accurate predictions.

Main Methods:

• Developed a DL-based framework, BindingSite-AugmentedDTA.
• Integrated a self-attention mechanism for interpretability.
• Augmented benchmark datasets with 3D protein structure information.
• Validated the framework computationally and experimentally.

Main Results:

• BindingSite-AugmentedDTA improved the performance of seven state-of-the-art DTA prediction algorithms.
• The framework demonstrated enhanced prediction accuracy and efficiency.
• Attention weights provided insights into protein binding sites, improving interpretability.
• Computational predictions showed high agreement with experimental results.

Conclusions:

• BindingSite-AugmentedDTA offers a generalizable and interpretable approach to DTA prediction.
• The framework has the potential to be a next-generation tool for drug repurposing.
• In-lab validation supports the practical applicability of the proposed method.